You! Yes, you, reading this… blog. A blog essentially unknown no less, with a name vaguely reminiscent of shapes and sounds. Let me ask you this:
You are conscious, aren't you? You take this for granted, this feeling, because it is familiar to you, this feeling of "being there". But have you ever wondered what it is that makes this feeling possible? How it is possible that you can have feelings at all? What feelings even are?
As scientists, we ask many questions about many difficult subjects. We wonder about the universe, the origin of life, and how to stop disease. Once in a while, scientists stop and wonder about their capacity to wonder. Yes, we are intelligent, in the sense that we can solve logical puzzles and perform computations in our head. (Yes, I mean you, and, well, ... some of us.) But solving puzzles and being able to compute is not a set of skills reserved solely to the primus of primates. Computers do this very well too. Sometimes better. They play chess and drive cars, make money on Wall Street, and predict the weather days in advance. But we give these machines nary a second thought. They are machines, and we built them. We do not care about them. Why would we? They do not care about us. Of course they don't, they do not have feelings. We care about animals, in part because we think they have feelings. While they don't compute the way we do (we believe), feelings are important to us. Having feelings makes all the difference. But we hardly ever ask where they come from.
This is a bit strange because, between you and me, you are a machine too. No matter how much you think that you are special because you are conscious, the difference between you with your feelings, and this machine in front of you, with a computer screen accurately relaying to you the words that I typed into another such machine not that long ago (but possibly very far away from where you read this) is precisely consciousness: the ability to feel something, to experience anything. Yet you never wonder why you have this ability (and your cat and your dog), but not that machine.
Can we even ask this question as scientists? Is consciousness, its origin, its construction, its evolutionary utility, fair game as an object of inquiry? Quite frankly, it hasn't been for the longest of times. As you can imagine, for those longest of times consciousness was synonymous with "soul", and any investigation of the origin and substance of the soul was the territory of that "other book", that also deals with origins but insists it has all the answers within 38,000 words, give or take. It took a Nobel laureate and his neuroscientist sidekick to make consciousness an object of serious scientific inquiry. Crick and Koch (yes, the famous Crick and the infamous Koch) famously asked: "If there is something in the brain that makes us conscious, we ought to be able to find which part it is. Show me the neural correlate of consciousness, and the mysteries surrounding it will be lifted".
But as the search for this neural correlate continues, it is slowly dawning on us that even if we find a particular place in our brain that turns consciousness on and off (should this place exist), we may still not understand how it works. For us to understand something, anything, don't we need to understand how consciousness works? Shouldn't we be able, for example, to make this "it" if we are to say that we understand it?
Can we make consciousness? Can we build conscious machines?
But as the search for this neural correlate continues, it is slowly dawning on us that even if we find a particular place in our brain that turns consciousness on and off (should this place exist), we may still not understand how it works. For us to understand something, anything, don't we need to understand how consciousness works? Shouldn't we be able, for example, to make this "it" if we are to say that we understand it?
Can we make consciousness? Can we build conscious machines?
"Hold your horses!", I can hear you through the digital divide, "how can you propose to make something that, er …, you do not, er,… know how to make"? "Easy, my friend", I reply. "We do this every day, we do this all the time!. We have at our disposal a process, an algorithm you may say, that can make things, even though we have absolutely not the faintest idea how they (the things we make) work."
"You've got to be kidding me!" I hear half of you say (while the other half goes: "I see where you're going with this!"). Yes, Darwinian evolution is a process that creates complex "things" without having to ask for blueprints, a business plan, a timeline to completion, and an estimate of market penetration within the next quarter. Evolution makes things that work, without theory, without understanding. (If you define by "evolution" things that ended up on the line of descent, rather than the myriad of failed attempts that ended up, well, not on the line of descent.) Evolution, most assuredly, makes things we don't understand. (Ask any biologist: if we understood all the things it produces we would not be blogging about biology. Or, reading blogs about biology. Or writing research papers. About biology.)
"All right, fair enough", you exclaim," biological evolution can do it, in fact, DID it, I stipulate. But can you?"
That is indeed question number one. (Question number two will follow in due time). Now you're asking me. Because I have a reputation of "evolving things". Yes, I have evolved things, even complex things. Have I evolved things that no human can design? That's a difficult question to answer unless you have a competition of sorts. Let me, at this point, simply say that this point may or may not be settled soon, and instead point to the bigger question: Can you use artificial evolution (by which I mean evolution within a computer) to evolve consciousness?
If you have been paying attention (and I have zero doubt that you have, otherwise THIS word would not have been read by you) there is a question burning in the back of your mind. "Granted everything", you commence, "granted that you are the grand voodoo of digital evolution, and that you can 'evolve complex things', as you ineloquently put it--this helps you squat because whatever you evolve, it will just be … something". "How will you prove to me and others that it is consciousness that you evolved, unless the "thing" walks out of the computer and says 'Cogito ergo sum'? (And speaks Latin for no apparent reason)."
That's a fair point. How can we claim that we evolved something that we don't know how to make? Don't we have to have a measure of that thing, so that we can say: "On the scale of the "Consciousness-O-meter" we achieved 0.8, so bite me and go away. We'll be at 0.9 next quarter!". Yes, if we had this, things would be so much easier. If only we had a way to measure that thing that we know not how to make. But clearly that's impossible. Or is it?
Can we measure something that we don't understand? (This is question number two, in case you are keeping track).
Once you turn this question over in your mind, you realize that this is a somewhat subtle question. For example, we were able to measure gravitational acceleration, say, before we understood gravity. (I shouldn't really say "we" in the same sense as I've use "we" here before. "Galileo was able to measure gravity before he understood gravity." There, that's better.) The latter is certainly true. But Galileo was lucky in a way. Gravity is so pervasive that its effect on our measurement devices is almost impossible to miss. But what is the effect of consciousness on a measurement device? And for that matter, being pervasive is not what makes a phenomenon easy to measure because consciousness, to us, is perhaps even more pervasive than gravity. Gravity makes things fall. Consciousness makes us experience things. But this experience is private, it occurs within the boundaries of our personality and no measurement device can measure its strength, unless it reaches deep within the entrails of our thoughts and dreams. The particular hue of our experience, the tint of our perception, those cannot be characterized, we are sure. Or can they?
Enter Giulio Tononi, a neuroscientist and sleep researcher at the University of Wisconsin. Giulio is the kind of guy who as a kid writes a letter to Karl Popper (the eminent philosopher who wrote extensively about the brain with John Eccles) whether he should devote his life to studying consciousness. Popper, incidentally, has been a hero of mine for a bunch of things, not the least for dissing Niels Bohr for his outrageous views on quantum mechanics. So Tononi gets back a nice letter from Popper, and then figures he better get an M.D. along with a Ph.D., so he can figure out how this thing we call consciousness comes and goes when we sleep. But if this was not enough, he decides he needs to understand what it is, mathematically speaking, the brain does when it does its thing. Now, an M.D. and even a Ph.D. in neurobiology doesn't even come close to qualify you to push the envelope in the mathematics of how the brain processes information. But Tononi, unfazed, developed the theory of integrated information processing that is now at the heart of the most ambitious attempt at capturing the peculiarity of the computer that is creating every word that I write. As opposed to the one that I write this in to.
What is this theory? Well, if I could explain it in a sentence, it wouldn't be much of a theory. As far as theories goes, it is pretty tame: it does not introduce new particles or forces, nor does it introduce a radical departure from our thinking about computation, for example. What it does, instead, is to focus our attention on the different ways that computation can be achieved. Yes, Tononi fully acknowledges that our brain is a computing device: it takes in signals from the outside, manipulates them inside the computer we call our brain, and then comes up with decisions. This is, in essence, not different from what every single computing device does that we humans have designed. What Tononi says is that the brain does it a little bit differently.
You see, when you or I design a computing device (suppose we are in that profession) we make sure that we know what every element of that computer does, and that we know when it will send what result on to the next module. Because this is the way we make sure that the device does precisely what we want it to do. We call this design. A designer makes sure that the thing works. We do this by isolating parts, and connecting these parts in such a way that we have complete control. What we don't realize when we design this way, is that we give up just about everything else in return for this predictability.
When you think about any object that you experience, do you experience its shape, color, smell and feel, separately? Do you remember how to use it--and the dreams you have had about it--independently from the shape color, smell, and touch? No you don't, they are all one, they are the experience of the object. But if if you design your computer to process every aspect of an object independently (and at different times) you rob this computer of experiencing this object. Realizing this, Tononi set out to develop the mathematics of information integration, to fashion a mathematical measure that distinguishes processes that integrate, from processes that don't. His quintessential non-integrating device is the digital camera's photoreceptor, that faithfully reflects the world in its millions of pixels, but integrates nothing. At the other extreme is our brain, that takes the input from our own photoreceptors, but then integrates the hell out of it and merges it with our other senses and memories, to create sensations. To create, ultimately, you.
After all this mathematics is done, we are left with a number that Tononi calls "Phi" (also the title of his most recent book), which characterize the capacity of any computing machinery to integrate information. What does this construction do for you? Right now, nothing of course. But imagine Tononi could record the activity patterns of your brain as you read this... shall we call it a blog? Phi could tell you if you are dreaming or in dreamless sleep, because your brain integrates information only during dreaming sleep. In the dreamless sort, you are unconscious, you have no experience at all. And as it so happens, Tononi is in a position to make precisely these types of measurements, as the director of a sleep laboratory. But this is not where the usefulness of Phi ends. If Phi can measure whether or not you are conscious, then shouldn't it be able to determine whether people in a comatose state are in a vegetative state without any consciousness at all, or else in a so-called locked-in state, fully conscious but unable to communicate this state to the outside world (think "The Diving Bell and the Butterfly")? There is now some evidence that Phi can distinguish between vegetative and conscious states but definitive proof will only be available when we can record from brains in more sophisticated ways than we can do today.
"Fine" you say, "I grant you that there may be a way to measure whether a computing machine integrates, or just reproduces, information". "Unless I'm caught in a coma, why should I care?"
And I tell you in return: "You have forgotten the first half of this blog post haven't you, where I insisted that evolution--the process that knows not what it designs--can do things that people can't do." Yes, that's right: evolution can create computational systems that integrate information at high levels, because evolution is not concerned with beautiful, predictable design. Evolution is messy, opportunistic, and unpredictable. Evolution takes what works and runs with it, whether it is neat or not, whether it adheres to design standard ISO9000 or not. More importantly, evolution creates designs that wastes as little as possible, reusing the same components over and over, and doing all these things at the same time. As a result, evolved computing systems integrate information. At a massive scale.
So, what makes natural computing machines such as brains interesting is information integration, and from what we know today, this level of integration cannot be achieved by design. What should we learn from this? Well, I think you beat me to it, this time. We should use evolution as the process that leads us toward the undesignable, toward the computer that integrates (for reasons of expediency only) to such an extent that the objects it perceives don't just evoke a reaction, they evoke an experience. There is a good reason we should expect experiences to be selected for: they allow the recipient of the experience to make better predictions about the future, and further the survival of the species that has these experiences.
So, where do we go from here? I think that we need to put together a team that understands evolution, as well as information integration, and works on one singular goal: to harness the power of evolution to create complex systems that defy engineering. Such an endeavor must be supplemented by empirical research wherever possible: research into brain anatomy, functional imaging, connectome mapping, adaptation, learning, and more. The reason for this is that theory and computation ought not to proceed in a vacuum: after all, we have the thing we want to make right here in front of us. We would be foolish if we did not try to learn from it as much as possible.
In my lab at Michigan State University, we are trying to push this envelope by evolving brains that are increasingly complex, by designing fitness landscapes that are increasingly complex. We do a lot of this work in collaboration with Giulio Tononi at the University of Wisconsin and Christof Koch at the Allen Institute for Brain Sciences. And we are trying to create a much larger team here at MSU so that we can integrate empirical approaches, behavioral biology, neuroscience, cognitive science, as well as psychology and decision-making into this effort. Just a month ago, we had a Symposium here at MSU where some of the "players" in this endeavor made an appearance. There I was reunited with Jeff Hawkins, who graciously accepted my invitation to speak, and updated us on the progress in the realization of his own dream. (Readers of this blog remember our meeting at the Google Fest, for the simple reason that my recollection of this meeting was the first post to the blog).
Other invited keynote speakers at the Symposium were Giulio Tononi, Daniel Wagenaar from Caltech, Ken Stanley from the University of Central Florida, and Mike Hawrylycz from the Allen Brain Institute. Local speakers were Dave Knoester and Arend Hintze from my lab, as well as Natalie Phillips who discussed reading Jane Austen inside an fMRI magnet, and Karim Oweiss who showed amazing videos of monkey brains interfacing with computers.
So this brings us full circle. You. Yes, you! Do you think we should give this a shot? Do you think we can do this? Do you want to answer with a 2008 presidential campaign slogan?
I leave you to reflect about the photograph of the left upper corner of the blackboard in Richard Feynman's office at Caltech, as it was after he died (that's what you saw about 5 minutes ago on this blog, give or take depending on your speed of reading). I think I do not have to explain the quote's relevance with respect to what we endeavor to do. If we cannot understand what we cannot build, then building it is our only path to understanding the brain, and ultimately what makes you you, and me me. It is not an easy path. It may be the hardest path of all. But as far as I can tell, it is the only path.
